HNHacker News
TopNewBestAskShowJobs

brianchu

2,341 karma · joined July 15, 2012

Looking for the next thing.

Worked on machine learning for medicine at Cardiogram. Studied CS at UC Berkeley. Former undergrad researcher and teaching assistant for Berkeley's ML class. Former intern at Twitter on Ads Analytics Infrastructure.

Email: bc@brianchu.com

github.com/bchu

twitter.com/brrrianchu

submissionscomments
brianchu··on Graph Isomorphism Algorithm Breaks 30-Year Impasse
Most CS programs that I am aware of don't require real analysis. I know Berkeley EECS doesn't, and it's a very rigorous program. Stanford too.

I plan on taking it, but only because it shows up in a few esoteric theoretical machine learning contexts.

brianchu··on Likelihood for Gaussian Processes
Honestly, the biggest thing holding back kernels is deep learning (unfortunately?).
brianchu··on Ask HN: What are the recruitment tools/plugins used by startups?
In terms of ATS's (Applicant Tracking Systems), I've heard lots of recruiters like lever.co
brianchu··on Who Y Combinator Companies Want
Bit of a nitpick, really, but you want a basis, not necessarily just a set of eigenvectors.
brianchu··on Who Y Combinator Companies Want
I'm a CS student and most students I know dislike whiteboarding too (and by extension many, if not most, YC startup employees do too). But I'm pretty okay at them now, having had lots of practice. The key advantage for students I think is the ability to have sustained practice over the course of a month (e.g. > 4 onsites over the course of a month), which most students can afford since their schedules are flexible. Also, I've recently had interviews at a few YC companies you've heard of, and they had me code on my laptop. I didn't whiteboard at all except for system design questions.

Reply to below: Most companies have me do at least one non-technical system design and/or talk-about-a-project interview.

brianchu··on Wall Street Talent Is Moving to Silicon Valley
There's a glaring error, which makes me doubt the credibility of the rest of the article. Mary Meeker joined KPCB as a partner, not an analyst (huge difference). And this was 5 years ago, so it's old news.
brianchu··on TensorFlow Benchmarks
According to: https://github.com/soumith/convnet-benchmarks, it's an NVIDIA TitanX (12GB GPU memory), which is pretty much the top of the line GPU for training neural nets. From my limited experience, most deep learning in research - including much of the state of the art - is done on single GPUs. (My guess is that if your model doesn't fit in 12GB, your model has way too many parameters to practically train anyway).
brianchu··on Y Combinator: the X Factor of tech
I would be very surprised if YC was doing any sophisticated machine learning on HN or even their portfolio. I'm sure they have nice Excel spreadsheets on their finances and portfolio metrics, maybe a linear regression or two.

This should be a law: we tend to overestimate the amount of sophisticated machine learning used in anything.

brianchu··on Apache Singa, a Distributed Deep Learning Platform
According to the paper, the parameters fit on one GPU (or at least that one GPU was able to train the model). It was just too slow, so they trained on 8 GPUs in parallel. But those GPUs were still on the same machine (one node, multiple GPUs).
brianchu··on Apache Singa, a Distributed Deep Learning Platform
It's only single node, multiple GPUs, though, which already exists in other frameworks.
brianchu··on Blowing the Whistle on the UC Berkeley Mathematics Department
Doubtful. I was a student during that time and I can't recall any math or CS (or STEM) classes that cancelled a class session because of the strike. The most that happened was some teaching assistants cancelled their section for one day.
brianchu··on Blowing the Whistle on the UC Berkeley Mathematics Department
I think the stakes are quite high. If he is truly a superior lecturer, at stake is the calculus education of over 800 students per year at one of the top schools in the nation, many of whom will be future scientists and engineers.
brianchu··on Oxford University Machine Learning Course
Here's a radical opinion: obsessing over frameworks is counterproductive; you most certainly should not be evaluating a class on a subject that is so math/concept-intensive by the class's choice of implementation coding framework. Especially since many of these frameworks are more alike than different, in the grand scheme of things.
brianchu··on Niantic Labs splitting from Google
Interesting hearsay on Reddit (via Twitter) https://twitter.com/MikeIsaac/status/631630506300448769, claiming it was spun off because it wasn't considered valuable.
brianchu··on Berkeley CS 61AS – Structure and Interpretation of Computer Programs, Self-Paced
We're discussing what freshmen in CS should be learning, not what sophomores/juniors/seniors should be learning. Getting people hooked on cool, albeit superficial, things they can build is the right way to get a broader range of people interested in CS (as opposed to just people who started programming before coming to college).
brianchu··on Berkeley CS 61AS – Structure and Interpretation of Computer Programs, Self-Paced
I think the grass is always greener on the other side. I started with SICP in Scheme (Racket) in high school (our teacher had taught at Berkeley over the summer, I think). I don't feel any special reverence for Scheme, and don't particularly feel either way about functional programming. It's just another tool in the toolbox and I don't feel like my smart peers who started with Python are at any disadvantage whatsoever.

I do think that another major advantage for Python is the fact that you can do cooler things with it, faster. CS has a serious funnel problem and the quicker we can get students to do cool things with CS (GUI stuff, web stuff), the better CS education will be.

brianchu··on Minister No More
I think this move serves to weaken the political cover for the Eurogroup refusing to continue negotiations with Greece (they can't say "we refuse to negotiate because of Yanis"). I'm starting to think that this was his plan all along, and that he and Tsipras are outplaying the Eurogroup right now.
brianchu··on Silicon Valley’s Youth Problem (2014)
Some of the best responses to this article are on Quora: https://www.quora.com/What-does-Silicon-Valley-think-of-the-...

I really hope you can see past the fact there is a login requirement. There is a lot of really great content on Quora.

brianchu··on Ambitious teenagers are flooding into San Francisco
I know a few of the people here. Before going to college I was definitely tempted to work at a startup and not to go to college. The tech industry in SF is overflowing with money, so I'm not surprised these people are making this choice (unexpectedly, it's often the small-but-VC-funded startup desperate for talent that pays more $$$ the bigcorps, often run by founders unable to qualify the talent). Also, in my experience studying CS (at least at Berkeley) is definitely harder than working in tech (compare doing psets and pulling all-nighters to doing working on things you're pretty good at, surrounded by every perk imaginable).

I did end up making the judgment that deeply studying CS in my 4 years at Berkeley would help me in the long run.

brianchu··on Project: A Simple Operating System
They're all in C now, using Pintos (I'm in the class right now; the switch happened last semester, AFAIK).
brianchu··on What's really behind Slack's crazy growth?
At UC Berkeley I'm seeing a lot of student groups switching to Slack. For example, several CS classes (including one of the largest ones, Data Structures) are using Slack for internal communication. A lot of student clubs (including one I'm involved with) are using Slack now.

So the idea that Slack is only popular because of enterprise sales doesn't really hold water in my opinion. It really is a great product. All of these groups heard about Slack through word of mouth.

brianchu··on Ask HN: Why is liquidation preference acceptable and prevalent?
Here's the argument I've heard: Let's say you found a company. A VC invests $1M for 20% of the company, for a $5M post-money valuation. Then you immediately sell the company for $3M (less than $5M). You make $2.4M. The VC essentially loses $400k. The 1x liquidation preference protects them from losing that $400k.
brianchu··on Dijkstra on Haskell and Java (2001)
FWIW, UC Berkeley's intro CS course sequence goes through the following languages in order: [Scratch (in an optional gentle-intro CS course)], Python (with an emphasis on using it for functional programming), a tiny bit of Scheme, a tiny bit of SQL (before they used a toy logic programming language), Java, C, MIPS assembly. I think this walks down the ladder of abstraction very nicely.
brianchu··on Let the Other 95% of Great Programmers In
In the long run. IIRC labor supply is a lagging indicator, so in the short run (e.g. 1-2 years) it would be a bit closer to a zero-sum game.
brianchu··on Let the Other 95% of Great Programmers In
The kind of training necessary for becoming exceptional is entirely possible, but not scaleable, thus it's not a practical solution to the problem. I think the example you give supports that - mentorship tends to evolve organically from companies you work at, the colleagues you work with, etc. It'd be difficult to imagine good mentorship being assigned, and succeeding, at-scale.

Even the processes for training competent engineers are only barely scaleable. Most competent engineers are still being trained through colleges. Even newly emerging forms of training don't scale that well - take coding bootcamps with limited class sizes and the low success rate of things like Codecademy.

brianchu··on Thoughts on Startup School
It is a matter of expectations. I've been the Startup School for three times now. I don't go for the talks - I go for the people. The talks certainly have an inspirational bent to them, and most of the info can be gleaned from HN, PG's essays, and a whole bunch of blog posts.

Every Startup School there are a few good talks (Andrew Mason's and Reid Hoffman's were two I liked), plus office hours, which are engaging - but again, these are all uploaded online. I'm not sure how office hours participants are selected, but doing that would be an obvious reason to attend.

The real reason to attend Startup School is to meet a fairly interesting, relatively accomplished, friendly crowd of folks. There's a business bent to the demographic - everyone is interested in startups, but most people I meet are technical. If you really want to, there's the chance to talk to some of the YC partners. Some of the speakers will stick around to answer questions. Last year one of the Airbnb cofounders stayed to answer questions for a while.

To get the most out of it you definitely need to introduce yourself to random people. You certainly shouldn't pull out a laptop. I never sit down with friends because sitting down next to a stranger is a great opportunity to introduce yourself.

Over the past three events I've met a few people I've kept in touch with; that alone makes it worthwhile. It also makes a lot more sense if you're local to the Bay Area - it's a great excuse to meet up with friends in the area.

brianchu··on Apple allows hot code push for JS downloaded and run by WebKit
This sounds great. It bears noting that this form of code push is routinely implemented in native apps too - I know at least that the Quora and Facebook iOS apps make liberal use of web views (and code push). Native apps can take the opposite hybrid approach - sprinkle in web views where needed inside a native codebase, as opposed to sprinkling in native plugins (and a native wrapper) in a web codebase.
brianchu··on A Gene That Makes You Need Less Sleep?
One proposed reason for the evolution of sleep is facilitating neural plasticity (growth and change). These studies merely show benefits in the short-term. An interesting hypothesis to test is if people with this mutation, over the long-term, have the same cognitive skills as those without, assuming both parties get the amount sleep they naturally need.
brianchu··on Delivery Startups Are Back Like It’s 1999
The problem is with taking Instacart's $3.99 delivery fee at face value. Last I checked (which was a year ago - they've probably lowered this) on top of that they charge 10-20% markup on each item, and they also pocket any savings from coupons / sales / bulk / rewards card discounts, which I imagine can run another 10%. It's in this situation where Instacart can plausibly meet your quoted $20 figure.
brianchu··on Show HN: Practical Guide to Mobile Apps Using Phonegap and Famo.us
For each native platform, there actually aren't many (viable) choices. Android has Android (framework) and Java. iOS has iOS (framework), and a choice between two very similar and interoperable languages, Objective-C and Swift.

If you decide to go HTML5 cross-platform you have Phonegap vs. other web wrappers (wrappers); Ionic, Famo.us, Sencha, hand-rolling-your-own (frameworks); and finally just JavaScript (language).

This is ignoring other cross-platform things like Xamarin (which consists of Xamarin the framework, C#, and still also the iOS or Android frameworks, depending on your targets), Titanium, etc.

← PreviousPage 3 of 10Next →